Study population
The study included consecutive patients referred for CMR at our center for “suspected myocarditis” as the primary clinical question between December 2002 and December 2015 [9]. In real-world clinical practice, diagnosis of myocarditis is challenged by a lack of reference standards including the use of noninvasive measures, serum biomarkers, and even tissue pathology from invasive biopsies [10]. We included consecutive patients with suspected myocarditis raised as a referral indication with presenting signs/symptoms from either one of 2 groups: 1) acute chest pain syndromes with symptom onset < 2 weeks before CMR; 2) subacute (onset ≥2 weeks) of dyspnea, signs of left ventricular (LV) dysfunction, or ventricular arrhythmias syncopal spells or abnormal electrocardiogram (ECG). Exclusion criteria included: 1) any evidence of coronary artery disease (CAD) by either previous documented medical history, any previous or newly detected relevant CAD in non-invasive or invasive imaging or subendocardial LGE consistent of CAD in a territory subtended by a coronary vessel 2) any evidence of hypertrophic cardiomyopathy, arrhythmogenic right ventricular (RV) cardiomyopathy, cardiac sarcoidosis, or cardiac amyloidosis; and 3) any evidence of Takotsubo cardiomyopathy, constrictive pericarditis, Loeffler endocarditis, ventricular noncompaction, cardiac tumor, pulmonary embolism, or severe valvular disease. Seven hundred forty-four patients were included. Fifty-nine (7.9%) patients were excluded based on CMR findings consistent with: myocardial infarction (N = 35), biopsy-proven cardiac amyloidosis (N = 6), ventricular non-compaction (N = 3), Takotsubo cardiomyopathy (N = 4), constrictive pericarditis (N = 2), cardiac sarcoidosis (N = 2), Loeffler endocarditis (N = 2), and 1 each for arrhythmogenic cardiomyopathy, hypertrophic cardiomyopathy, pulmonary embolism, cardiac tumor, and severe valvular disease. Fifteen patients were excluded due to technical reasons (claustrophobia with incomplete CMR scans or non-diagnostic LGE images) [9]. Takotsubo cardiomyopathy was defined as previously published with apical ballooning [11], elevated troponin, absence of CMR features suggesting of myocarditis (absence of LGE) and absence of coronary artery disease. Clinical data including medication, laboratory tests including cardiac biomarkers, and ECG before CMR scanning were analyzed. Abnormal troponin I was defined < 0.10 ng/mL and troponin T < 14 ng/l. Normal values of creatine phosphokinase (CPK) were < 145 U/l in women and < 170 U/l in men. Clinical data, cardiac biomarkers and ECG were analyzed at baseline by a cardiologist.
CMR imaging protocol and image post-processing
The CMR systems included a 3T (Tim Trio, Siemens Healthineers, Erlangen, Germany) and a 1.5 T (Aera, Siemens Healthineers). All patients underwent cine balanced steady-state free precession (bSSFP) imaging and an LGE imaging protocol (TR, 4.8 ms; TE, 1.3 ms; inversion time, 200 to 300 ms), using a segmented inversion-recovery pulse sequence starting 10 to 15 min after a weight-based injection (cumulative dose 0.15 mmol/Kg) of gadolinium diethylenetriamine pentaacetic acid (Magnevist, Bayer HealthCare Pharmaceuticals Inc., Berlin, Germany). In 122 (21%) patients, Multihance (Bracco Diagnostic, Milan, Italy) was used instead of Magnevist. In patients with an estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73m2, contrast dose was restricted to 0.1 mmol/kg or 20 ml, whichever was lower in volume in compliance with our institutional policy [12]. A commercially available software (MASS v15, Medis, Leiden, The Netherlands) was used to post-process and quantify all CMR images.
LGE quantification
Semi-automated quantification was performed as follows: Epicardial and endocardial LV contours were carefully placed manually on all LGE images. The remote non-LGE reference region of each LGE slice was placed adjacent to the region of LGE so that the reference region is at approximate equal distance from the anterior receiver coils. Therefore, we believe this method minimizes any modifying effect from LGE location to robustness of the LGE quantitation. LGE mass was then quantified by semi-automatic methods using a signal intensity threshold of > 2,3,4,5,6,7-SD, respectively above a reference region of remote myocardium (adjacent to the region of LGE and approximately equal in distance to anterior receiver coils) in the same slice, and using regions defined as above 50% of maximal signal intensity of the enhanced area for the FWHM approach (see Fig. 1) [13, 14]. Artifacts were manually erased. In all methods LGE mass (in grams), was then expressed as a percentage of total LV mass determined by bSSFP cine images [15]. LGE extent was also determined visually by the 17 segments model by using two different scores: [1] LGE being present or not in the segment defined the visual presence score (LGE-VPS) (maximum score 17), and [2] the visual transmurality score (LGE-VTS) summed the transmural extent of LGE per segment, assessed by a five-point scale (0 = no LGE, 1 = < 25% transmurality, 2 = 26–50% transmurality, 3 = 51–75% transmurality, 4 = 76–100% transmurality) (maximum score 68). VPS sizing aimed to include any abnormal enhancement on LGE images including regions of intermediate signal intensity. LGE image quality was graded as: 1 = poor image quality, 2 = fair image quality, 3 = good image quality, and 4 = excellent image quality. LGE images were evaluated by the consensus of two American College of Cardiology Core Cardiovascular Training Statement (COCATS) level III experienced readers (CG and RYK) and inter-rater reproducibility testing was performed by an independent experienced CMR investigator (KK).
Follow-up of clinical endpoints
We reviewed all available electronic medical records including mortality status from Social Security Death Index for all subjects. Subjects were then sent a standardized checklist-based patient-questionnaire by mail and/or followed-up by a scripted telephone interview based on the same standardized checklist 2 weeks later. Subjects were given an option to refuse to be contacted by telephone whether or not they decided to return the questionnaire by mail. Subjects who we were not able to establish contact for at least 6 months after the CMR were considered lost to follow-up. We defined the primary endpoint being a composite major adverse cardiac events (MACE): a) all-cause death, heart failure decompensation requiring hospital admission as defined in prior trials [16, 17], heart transplantation, documented sustained ventricular arrhythmia (> 30 s), recurrent acute myocarditis based on clinical biomarkers and CMR Lake-Louise definition [6]. Time-to-MACE was determined from the CMR study date to the occurrence of the first MACE or censorship at end of the follow-up period. All study procedures were reviewed and approved by our Institutional Review Board in accordance with institutional guidelines.
Statistical analysis
Categorical variables were presented as percentages of the entire cohort or as percentage of the corresponding group if relevant data were missing. Continuous variables were expressed as mean ± standard deviation or as median values with interquartile range [IQR] depending on normality of distributions. Categorical variables were compared using the Chi2 or Fisher exact test in low field numbers, whereas comparisons for continuous data were performed using 2-sample Student t-test or Wilcoxon rank-sum test, when appropriate. A two-sided p-value of < 0.05 was deemed significant. Univariable associations with MACE and chi-square were determined by Cox proportional hazards regression. Multivariable (simultaneous entry, i.e. enter method) associations of risk covariates with MACE were determined by Cox proportional hazards regression. Multivariable models were built by including LGE extent from the various quantification methods and clinical variables in order to test for the prognostic association of LGE extent incremental to common clinical risk markers. For each multivariable model, the assumption of proportional hazards was checked and confirmed valid. For inter-rater and intra-rater reliability analyses, we randomly selected 20 patients with LGE presence and compared the measurement of LGE extent using all 7 quantitative methods (2-SD through 7-SD, FWHM) by two independent readers. Each CMR scan was independently contoured by the 2 readers and automated region of reference (ROI) was selected to generate the LGE extents for comparison. Furthermore, the LGE-VPS and LGE-VTS visual scores by two independent readers were assessed for Intra- and inter-reader agreement using intraclass-correlation coefficient (ICC). SAS (version 9.4, SAS Institute Inc., Cary, USA) and SPSS 22.0 (International Business Machines Corporation, Armonk, New York, USA) software packages were used for analysis.
